Agents Are Amazing but Still Frustrating

There’s something missing with every agent I’ve used. @bot is no different.

The first few days it all seems amazing, surprising, revolutionary. Your expectations grow. Kind of like FSD; it’s so good, that the edge case mistakes feel so damaging.

As you delve deeper with agents by asking more questions, and handing over more complex tasks, the cracks begin to show.

The lack of initiative and intuition.

The lack of context and memory.

The lack of insight and judgement.

The flip-flopping and loose opinions that erode your trust in their outputs.

It begins to get frustrating. You constantly have to poke, prod and remind your agent to keep them on task. It’s like having this incredibly intelligent employee, with zero agency.

We’re just not there yet. They’re amazing, but frustrating.

The average user doesn’t want to set up routines, automations and plugins. They just want it to work. The promise of ai agents is so great, so tantalising, that we’re willing to forgive so much.

It’s a precarious position for Anthropic and OpenAI to be in. People are willing to jump ship to a different product because no model or agent is so good yet that it’s impossible to leave. No one has ecosystem lock in.